Particle.news
Download on the App Store

Google Sets Ironwood TPU for Public Launch and Rolls Out Axion CPU Cloud Instances

The push underscores a vertical strategy to lure AI workloads with custom chips that scale to massive clusters.

Overview

  • Ironwood, Google’s seventh‑generation TPU, will be available to customers in the coming weeks after internal testing that began in April.
  • Ironwood pods can link up to 9,216 chips and slot into Google’s AI Hypercomputer design, which uses optical circuit switching to form large, reconfigurable clusters.
  • Google cites major performance gains over prior TPUs and disclosed specs such as 4.6 petaFLOPS FP8, 192 GB of HBM3e, and 7.4 TB/s bandwidth, placing Ironwood in the performance range of recent Nvidia parts.
  • Anthropic plans to access up to one million Ironwood TPUs for Claude, with other customers beginning deployments on the new platform.
  • Alongside the accelerator, Google introduced Axion Armv9-based instances—C4A generally available and N4A plus C4A Metal in preview—as the company ramps investment to meet strong AI infrastructure demand.